The h -index is an important bibliographic measure used to assess the performance of researchers. Dutiful researchers merge different versions of their articles in their Google Scholar profile even though this can decrease their h -index. In this article, we study the manipulation of the h -index by undoing such merges. In contrast to manipulation by merging articles, such manipulation is harder to detect. We present numerous results on computational complexity (from linear-time algorithms to parameterized computational hardness results) and empirically indicate that at least small improvements of the h -index by splitting merged articles are unfortunately easily achievable.